app-shell metadata-admin inspectors escape to the real network via useObjectFields/useObjectOptions (ECONNREFUSED noise, different probe than #4688) #4697

Activity

1. yinlianghui

   added

   pm:dispatched

   finding

   on Aug 15, 2026
2. yinlianghui

   self-assigned this

   on Aug 15, 2026
3. yinlianghui commented on Aug 15, 2026

   

   yinlianghui

   on Aug 15, 2026 – with Claude

   ContributorAuthor

   More actions

   First-touch grading + claim (seat repo:objectui; session session_01RnQd8iMMUwXQEV1crFmQiQ, 2026-08-15): promoted and dispatched in one stroke — pm:dispatched, size S, tests-only. The evidence is complete (per-file counts summing to the measured 180, URL proof at the fetch call site, bisection excluding every other directory) and the remedy has an in-package precedent (FlowReferenceField.lookup.test.tsx's vi.mock('../useMetadata') stub client shape). Acceptance = ECONNREFUSED 180 → 0 across the 9 named files with all assertions unchanged.

   Claim: PM loop round 11
   Session: session_01RnQd8iMMUwXQEV1crFmQiQ
   Branch: claude/issue-4697-metadata-client-doubles
   Worktree: objectui-issue-4697
   Domain: repo:objectui
   File surface: the 9 named test files under packages/app-shell/src/views/metadata-admin/inspectors/ (+ a shared local stub helper if the repo's in-package convention allows) + the tests-only empty-frontmatter changeset if the presence gate demands one. ⛔ No production source anywhere (stop on breach; explain in the report)
   Container & model: S, mode:subagent, model: sonnet
   Serial constraints cleared: in-flight is only objectstack#8830 (other repo); PR #4698 (plugin-view half) is in CI on a disjoint package; no open dev PR touches these files (checked this round).

   ---

   Generated by Claude Code
4. yinlianghui commented on Aug 15, 2026

   

   yinlianghui

   on Aug 15, 2026 – with Claude

   ContributorAuthor

   More actions

   <!-- os-dev-report -->

   {
     "issue": 4697,
     "status": "done",
     "branch": "claude/issue-4697-metadata-client-doubles",
     "pr": "https://github.com/objectstack-ai/objectui/pull/4700",
     "premise_still_valid": true,
     "summary": "Reproduced the card's exact baselines on current main (e05db8826): 70/40/12/12/12/12/8/8/6 ECONNREFUSED per file, sum 180. Stubbed useMetadataClient (../useMetadata) in each of the 9 named inspector test files with a per-file vi.hoisted()+vi.mock stable stub client ({get,list} both resolving empty), mirroring FlowReferenceField.lookup.test.tsx's precedent -- no shared helper file, since 9 short self-contained blocks reviewed more clearly than a new cross-file abstraction. Every file re-run isolated (maxWorkers=1) now shows 0 ECONNREFUSED with every existing assertion passing unchanged (same pass counts: 22,9,6,6,4,2,4,4,3 = 60 tests). Added a tests-only empty-frontmatter changeset (check-changeset-presence.mjs demanded one).",
     "tests": "Baselines on main (e05db8826), each file isolated via `pnpm exec vitest run` with maxWorkers=1: PageBlockInspector.i18n=70, .sectionName=40, ConditionBuilder.celGate=12, HookDefaultInspector.condition=12, ViewVariantInspector.celGate=12, .homeGate=12, ActionDefaultInspector.celGate=8, ConditionBuilder.test=8, HookDefaultInspector.celGate=6 (sum 180, matches the card). After the fix, same isolated commands: all 9 files 0 ECONNREFUSED, all previously-passing assertions still pass (22/22, 9/9, 6/6, 6/6, 4/4, 2/2, 4/4, 4/4, 3/3). Directory suites: the inspectors/ directory -- 53 files, 590 passed plus 1 skipped (pre-existing, unrelated), 0 ECONNREFUSED. Full metadata-admin/ directory -- 167 files, 1671 passed plus 1 skipped, 0 ECONNREFUSED (a separate ECONNRESET noise from previews/FlowCanvas.test.tsx, unrelated mechanism, filed as #4699). Reverse verification (direction predicted first: revert then count returns to baseline, tests stay green): committed the fix (b65801b2d), reverted PageBlockInspector.i18n.test.tsx to the pre-fix blob via git checkout, re-ran isolated -- 70 ECONNREFUSED, 22/22 still passing (confirmed prediction); restored the file from the fix commit -- back to 0 ECONNREFUSED, 22/22 passing. Gate union at HEAD b65801b2d (quoted from the post-commit run): dependency closure build (pnpm --filter with the app-shell-dependents pattern) green; app-shell's type-check script (tsc --noEmit plus tsc -p tsconfig.test.json) green; eslint --quiet on the 9 changed files clean, no output; node scripts/check-changeset-presence.mjs green (1 empty-frontmatter changeset declared); pnpm run check:control-bytes green (4223 files scanned). scripts/pm/dispatch-gates.mjs does not exist in this repo (objectui, unlike objectstack) so I hand-derived implicated gate families against the actual diff (9 test files plus 1 changeset): no new error codes, no .claude/agents changes, no new fake engine, no t() call-site changes, no doc changes -- no additional family beyond the ones named/run above.",
     "open_questions": [],
     "out_of_scope_findings": [
       "filed as #4699: app-shell previews/FlowCanvas.test.tsx escapes to the real network (ECONNRESET / socket-hang-up noise, a different transport mechanism than #4688's/#4697's plain-fetch ECONNREFUSED) -- 19 lines isolated, harmless (all assertions pass), found while re-running the full metadata-admin/ suite for this card's own step-2 verification; unassigned, not touched here"
     ]
   }

   ---

   Generated by Claude Code
5. yinlianghui commented on Aug 15, 2026

   

   yinlianghui

   on Aug 15, 2026 – with Claude

   ContributorAuthor

   More actions

   os-dev-report

   (Correction: the previous comment's <!-- os-dev-report --> marker survived only as escaped entities, not as a literal match. This comment is the report of record.)

   {
     "issue": 4697,
     "status": "done",
     "branch": "claude/issue-4697-metadata-client-doubles",
     "pr": "https://github.com/objectstack-ai/objectui/pull/4700",
     "premise_still_valid": true,
     "summary": "Reproduced the card's exact baselines on current main (e05db8826): 70/40/12/12/12/12/8/8/6 ECONNREFUSED per file, sum 180. Stubbed useMetadataClient (../useMetadata) in each of the 9 named inspector test files with a per-file vi.hoisted()+vi.mock stable stub client ({get,list} both resolving empty), mirroring FlowReferenceField.lookup.test.tsx's precedent -- no shared helper file, since 9 short self-contained blocks reviewed more clearly than a new cross-file abstraction. Every file re-run isolated (maxWorkers=1) now shows 0 ECONNREFUSED with every existing assertion passing unchanged (same pass counts: 22,9,6,6,4,2,4,4,3 = 60 tests). Added a tests-only empty-frontmatter changeset (check-changeset-presence.mjs demanded one).",
     "tests": "Baselines on main (e05db8826), each file isolated via `pnpm exec vitest run` with maxWorkers=1: PageBlockInspector.i18n=70, .sectionName=40, ConditionBuilder.celGate=12, HookDefaultInspector.condition=12, ViewVariantInspector.celGate=12, .homeGate=12, ActionDefaultInspector.celGate=8, ConditionBuilder.test=8, HookDefaultInspector.celGate=6 (sum 180, matches the card). After the fix, same isolated commands: all 9 files 0 ECONNREFUSED, all previously-passing assertions still pass (22/22, 9/9, 6/6, 6/6, 4/4, 2/2, 4/4, 4/4, 3/3). Directory suites: the inspectors/ directory -- 53 files, 590 passed plus 1 skipped (pre-existing, unrelated), 0 ECONNREFUSED. Full metadata-admin/ directory -- 167 files, 1671 passed plus 1 skipped, 0 ECONNREFUSED (a separate ECONNRESET noise from previews/FlowCanvas.test.tsx, unrelated mechanism, filed as #4699). Reverse verification (direction predicted first: revert then count returns to baseline, tests stay green): committed the fix (b65801b2d), reverted PageBlockInspector.i18n.test.tsx to the pre-fix blob via git checkout, re-ran isolated -- 70 ECONNREFUSED, 22/22 still passing (confirmed prediction); restored the file from the fix commit -- back to 0 ECONNREFUSED, 22/22 passing. Gate union at HEAD b65801b2d (quoted from the post-commit run): dependency closure build (pnpm --filter with the app-shell-dependents pattern) green; app-shell's type-check script (tsc --noEmit plus tsc -p tsconfig.test.json) green; eslint --quiet on the 9 changed files clean, no output; node scripts/check-changeset-presence.mjs green (1 empty-frontmatter changeset declared); pnpm run check:control-bytes green (4223 files scanned). scripts/pm/dispatch-gates.mjs does not exist in this repo (objectui, unlike objectstack) so I hand-derived implicated gate families against the actual diff (9 test files plus 1 changeset): no new error codes, no .claude/agents changes, no new fake engine, no t() call-site changes, no doc changes -- no additional family beyond the ones named/run above.",
     "open_questions": [],
     "out_of_scope_findings": [
       "filed as #4699: app-shell previews/FlowCanvas.test.tsx escapes to the real network (ECONNRESET / socket-hang-up noise, a different transport mechanism than #4688's/#4697's plain-fetch ECONNREFUSED) -- 19 lines isolated, harmless (all assertions pass), found while re-running the full metadata-admin/ suite for this card's own step-2 verification; unassigned, not touched here"
     ]
   }

   ---

   Generated by Claude Code
6. yinlianghui commented on Aug 15, 2026

   

   yinlianghui

   on Aug 15, 2026 – with Claude

   ContributorAuthor

   More actions

   ACCEPT — PR #4700 (reviewer of record; readings taken on GitHub at head b65801b2d, matching the report's gate-union HEAD).

   • Shape: draft → main, first line Fixes #4697 — full deliverable; app-shell previews/FlowCanvas.test.tsx escapes to the real network (ECONNRESET/"socket hang up" noise, different mechanism than #4688/#4697) #4699 referenced bare. Scope (10 files): the 9 named test files + the tests-only empty-frontmatter changeset the presence gate demanded — no production source, exactly the boundary.
   • Acceptance met as measured: per-file baselines reproduced exactly (70/40/12/12/12/12/8/8/6 = 180), all to 0 isolated, with every assertion count unchanged (60 tests). The directory and full-area suites confirm nothing else moved. Reverse verification on the biggest file returned exactly to its 70-line baseline with tests green — proving the stub, not something else, silences the noise.
   • Mechanism quality: the stable-identity stub reasoning (the hooks' effects key on [client, …], so a fresh object per render would re-fire them) is the detail that separates a correct double from a subtly-flaky one; the no-shared-helper call is fine at this size.
   • Out-of-scope finding: app-shell previews/FlowCanvas.test.tsx escapes to the real network (ECONNRESET/"socket hang up" noise, different mechanism than #4688/#4697) #4699 (FlowCanvas ECONNRESET — a different transport mechanism, 19 lines) — correctly filed, not folded; graded on its own card.

   CI on this head is converging — ready + auto-merge flips when the gates conclude green (timer armed). Card closes on merge; pm:dispatched comes off then. With this and PRs #4689/#4698, the entire measured ECONNREFUSED noise across the four packages is attributed and closed.

   ---

   Generated by Claude Code
7. yinlianghui

   closed this as completedin #4700on Aug 15, 2026
8. yinlianghui

   removed

   pm:dispatched

   on Aug 15, 2026
9. akarma-synetal

   added 2 commits that reference this issue on Aug 17, 2026

   test(app-shell): stub useMetadataClient in metadata-admin inspector s…

   ...

   This issue will close once commit 01c3e76 is merged into the 'main' branch.

   01c3e76

   test(app-shell): stub the automation/actions fetch in FlowCanvas.test…

   ...

   This issue will close once commit 03e98fc is merged into the 'main' branch.

   03e98fc